The direct answer
For the selected PostgreSQL profile, use the exact transaction-level try-lock result only within an admitted, exclusively owned transaction and session. Treat existing same-session lock state as a separate prerequisite, because a successful request is not automatically a fresh per-job claim. Advisory coordination does not authorize row access, establish commit success or protect an external action.
Choose one bounded database-local operation
A fictional AI workspace maintains one internal metadata marker R-A for an already admitted document revision. Two application requests might reach the marker’s maintenance path together. The model supplies no SQL or lock key; trusted code performs only a fixed bounded database-local operation after current authorization.
The reader’s question is narrower than whether the database account may edit R-A: which lock operation and lifetime support entry to this cooperative critical section? The existing row-policy guide answers a different old-row/new-row authority question.
This worksheet selects PostgreSQL 18 documentation and one direct scalar pg_try_advisory_xact_lock(bigint) operation. It uses symbolic attempt A-A, session S-A, transaction T-A and trusted key K-A. No installed database, patch release, client or pool was inspected. Every card is PROPOSED_UNRUN: no SQL, connection, transaction, lock or catalog operation occurred.
Freeze the function and interpret its actual result
The advisory-function reference describes pg_try_advisory_xact_lock(bigint) as an exclusive transaction-level request. It returns true on immediate acquisition, or false without waiting when acquisition is unavailable. Its transaction-level lifetime ends with the current transaction; there is no manual release.
Our proposed application accepts only the selected function and trusted bigint key. There is no relation scan, shared-lock branch, two-integer alternative, caller-selected key or extra SQL. A false result means this attempt does not enter. The worksheet supplies no spin, delayed retry, new-key escape or weaker function substitution.
An error, missing result or uncertain transaction association is neither accepted true nor ordinary contention false. Keep that state unresolved. “Without waiting” describes this lock-acquisition choice, not a bound on client round trips, query execution, application latency or all other database work.
Separate transaction lifetime from session lifetime
PostgreSQL’s locking chapter distinguishes session locks, held until explicit release or session end, from transaction locks released at transaction end. A session lock acquired inside a rolled-back transaction remains held. Session and transaction requests for the same identifier conflict across sessions, but additional requests by a session already holding that lock succeed across both levels.
That last rule changes the admission decision. A-A cannot infer a new independent job claim merely because S-A returned success. Another logical task, an earlier session-level hold or unexpected sharing can invalidate the application’s ownership story even though the database follows its documented rules.
Require independently supported evidence that S-A belongs exclusively to A-A, has no relevant earlier lock state and keeps T-A’s actual boundary through the protected work. Those prerequisites are unimplemented here. Unknown state holds the plan; an earlier checkout record is not a lock-state or ownership proof. This article supplies no pool reset, manual unlock or client-disposal recipe.
Make the key and every participant part of the contract
The same chapter calls these locks advisory because PostgreSQL does not enforce application participation. K-A therefore has an application-defined meaning. Require one reviewed, stable association between R-A and K-A, and require every relevant participant to follow the same scope. The model cannot choose a different key to avoid contention.
The proposed admission record binds attempt, resource revision, key-mapping revision, exact function, session owner and transaction context. It separately records current row/action authority and local work bounds. A hash or a familiar integer is not evidence that the mapping is collision-free or consistently used; the actual design needs review.
The conditional positive permits the fixed metadata critical section only when those prerequisites and the selected true result hold together. It does not report that the metadata changed. Supported completion still needs the application’s actual transaction outcome. No model call, message, payment, storage request or other external operation belongs inside this worksheet’s protected scope.
Walk eight inert scope-admission cards
| Invented condition | Proposed disposition |
|---|---|
| Clean exclusive S-A/T-A; correct K-A/R-A mapping; selected result true; all other prerequisites hold | Admit the bounded cooperative local section. Record no executed or committed change. |
| The selected result is false | Do not enter. Report contention without an automatic loop or another key. |
| A session-level function replaces the selected function, with rollback offered as release proof | Reject the changed lifetime; rollback is not the declared session-lock release evidence. |
| Work moves to another session or continues after T-A ends | No transfer of this reviewed scope; it is not a lease for later work. |
| S-A already holds relevant state, or another logical job shares it | Hold the per-attempt claim; success is insufficient fresh-ownership evidence. |
| The same resource uses another mapping or a caller-selected key | Reject coordination drift rather than infer mutual exclusion from different names. |
| A relevant writer bypasses the advisory protocol | Do not claim universal serialization or a database permission boundary. |
| Lock admission matches, but current rights, transaction outcome or local-only bounds are unknown | Withhold work or preserve unknown outcome as appropriate; the lock settles neither authority nor commit. |
A future reviewer could walk these configuration cards without touching a database. This article has not run that exercise or measured contention. Keep minimized attempt and profile references rather than row contents, connection details or credentials in ordinary evidence.
Keep coordination smaller than an execution guarantee
The complete positive also requires a trusted fixed adapter, supported actual transaction handling, bounded local work and the independently reviewed rights profile. No savepoint or exception-block scope changes are selected. None of those implementation requirements is demonstrated by this documentation review.
The existing reporting profile covers a different connection and permission design. The notification guide covers discovery hints, not durable claims. Advisory entry here grants no job lease, external fencing, exactly-once effect, snapshot freshness, fairness, deadlock freedom or resource ceiling.
No SQL, function, session, client, database, runtime or model experiment occurred. Boundary is prelaunch and supplies no shipped PostgreSQL coordinator or permission engine. Its synthetic policy evidence concerns different behavior.
Primary sources
Official references supporting the guidance. Examples in this guide are synthetic and are not reports of a production incident.